Scaffolded Vulnerability: Chatbot-Mediated Reciprocal Self-Disclosure and Need-Supportive Interaction in Couples

TL;DR

This work investigates how a chatbot can scaffold high-quality self-disclosure between couples by grounding design in Self-Determination Theory (SDT). It introduces a dual-layer scaffolding framework: Layer 1 provides enabling instrumental support (autonomy, competence, relatedness) to create a safe space for disclosure, and Layer 2 offers mediating relational prompts that guide partners to provide autonomy-, competence-, and relatedness-support to one another. In a randomized study with $N=72$ individuals (36 dyads) across three conditions—Partner Support (PS), Direct Support (DS), and Basic Prompt (BP)—the authors show that enabling affordances deepen disclosure while mediating affordances robustly elicit partner-provided need support and enhance perceived closeness; vitality and other well-being indicators also improve under scaffolded conditions. The findings offer empirical support for SDT-guided mediation as a mechanism to foster genuine connection in AI-mediated conversations and provide concrete design implications for embedding volume, pacing, and reflection prompts into everyday messaging platforms. Overall, the study contributes an SDT-based design framework—Dual-Layer Scaffolding—that enables AI to facilitate, rather than replace, human intimacy by nurturing autonomous, competent, and connected interactions within couples.

Abstract

While reciprocal self-disclosure drives intimacy, digital tools seldom scaffold autonomy, competence, and relatedness -- the motivational underpinnings defined by Self-Determination Theory (SDT) that enable deep exchange. We introduce a chatbot employing dual-layer scaffolding to satisfy these needs: first providing enabling affordances (instrumental support) for vulnerability, then mediating affordances (relational support) for responsiveness. In a randomized study (N = 72; 36 couples) comparing Partner Support (PS: both layers), Direct Support (DS: enabling only), and Basic Prompt (BP: questions only), results reveal a critical distinction. While enabling affordances (PS, DS) were sufficient to deepen disclosure, only mediating affordances (PS) reliably elicited partner-provided need support and increased perceived closeness. Furthermore, controlled motivation decreased across conditions, and scaffolding buffered vitality, which remained stagnant in BP. We contribute empirical evidence that SDT-guided mediation fosters connection, offering a practical framework for designing AI-mediated conversations that support, rather than replace, human intimacy.

Figures (11)

• Figure 1: Example workflow of the chatbot's conversation system in Phase 4 (Competence Question). (1) After being called in the Telegram group chat, the Driver LLM (II) receives input consisting of the current chat log (I), general prompts (III), phase-specific prompts for the active phase (IV), and a structured summary of prior interactions (V). (2) Based on these inputs, the Driver LLM generates a response (VI) and delivers it back to the group chat. (3) When the phase transitions, the Analyzer LLM (VII) processes the completed chat log and produces a structured summary (VIII), which is added to the conversation context for subsequent phases.
• Figure 2: Experiment flow. Participants first completed a sign-up form and a pre-interaction survey assessing demographic information, interpersonal closeness, self-esteem, vitality, positive affect, motivations for relational activities, and need satisfaction. Dyads were then randomly assigned to one of three experimental conditions, PS, DS, or BP, and engaged in a chatbot-mediated conversation following the condition-specific structure. Conversation data were subsequently analyzed using both quantitative measures (chat duration, number of messages, number of words) and qualitative coding (self-disclosure, basic psychological need support). Finally, participants completed a post-interaction survey and a semi-structured interview.
• Figure 3: Per-group phase timelines across the three conditions (minutes). Each bar (G1–G36) shows the duration of conversational phases. Partner Support (PS, G1–G12) includes all seven phases; Direct Support (DS, G13–G24) includes rapport and selected questions; Baseline (BP, G25–G36) includes only the question phases.
• Figure 4: Estimated marginal means of interpersonal closeness (IOS), self-esteem, vitality, and positive affect before and after the interaction. Error bars represent 95% Confidence Intervals..
• Figure 5: Perceived need support from the chatbot across conditions.